U.S. Cybersecurity Agency CISA Deploys Anthropic's Mythos AI Model to Audit Government Code for Security Vulnerabilities

According to a Reuters report, the U.S. Cybersecurity and Infrastructure Security Agency (CISA) has begun using the Mythos large language model (LLM) developed by AI startup Anthropic to conduct large-scale security audits of government code repositories, proactively identifying potential cybersecurity vulnerabilities.

Attack Surface Evaluation Team Leads Scanning Operations

Sources familiar with the matter say the scans are being carried out by CISA's internal Attack Surface Evaluation Team. This dedicated unit has long conducted digital security assessments and simulated cyberattack exercises across federal agencies, and serves as a core component of CISA's active defense strategy.

Significant Number of Vulnerabilities Already Discovered

Multiple sources indicate that the audits conducted using the Mythos model have already yielded preliminary results, uncovering a substantial number of code-level vulnerabilities. This development signals that the use of generative AI for government cybersecurity has moved beyond proof-of-concept and into active operational deployment.

Implications of AI-Assisted Security Auditing

The collaboration represents a significant strategic shift for the U.S. federal government—moving away from traditional manual code reviews or conventional automated tools toward using large language models for code-level vulnerability analysis. Anthropic's Mythos model is reported to possess a deep understanding of code semantics and logical structure, enabling it to identify complex vulnerabilities that traditional static analysis tools may struggle to detect.

Neither CISA nor Anthropic has publicly commented on the report. As AI adoption in government cybersecurity continues to expand, concerns remain regarding data privacy, model reliability, and the processes by which audit findings are acted upon.

Sources: Reuters / Slashdot

原文來源： 查看原文